Perplexity’s 2026 user count is reported as 30 million, 34 million, 45 million, 100 million and 230 million. Every one of those figures appears in articles published within the same twelve months. The spread is 7.7x.
The single most-quoted data point in the entire Perplexity corpus contains a 19% internal contradiction that no secondary source has flagged. A target the CEO described as conditional has been republished as an achievement for a year.
This report separates what the data measures from what the citation chain has turned it into.
Every figure verified against the primary source. No aggregators, no secondary summaries, no AI fact-checkers. Derived percentages are shown with their inputs.
The reported range
Seven figures are in active circulation for the same company in the same year. Three of them are not user counts at all.
| Figure | What it counts | Date |
|---|---|---|
| 30 million MAU | Core product only. Company-stated, most-cited benchmark. | Apr 2025 |
| 34 million MAU | Unspecified scope. Secondary compilation. | Mid 2026 |
| 45 million MAU | Unspecified scope. Most-repeated 2026 figure. | Early 2026 |
| 100 million MAU | All products combined. Sacra, citing Srinivas. | Apr 2026 |
| 230 million MAU | Unverifiable. GEO agency marketing page. | Q1 2026 |
| 138.1 million | Monthly website VISITS, not users. Similarweb. | May 2026 |
| 80.5 million | Lifetime app DOWNLOADS, not users. | Late 2025 |
Visits are sessions. Downloads are installs. Neither is a person, and neither is monthly.
Excluding the unverifiable 230 million claim, the credible range runs from 30 million to 100 million. That is 3.3x. The gap is not disagreement about facts. It is the difference between counting one product and counting everything the company ships.
Figure 1. Published Perplexity user figures, 2025 to 2026, by what each one counts.
Fracture 1: The anchor figure does not reconcile with itself
Every serious Perplexity statistic in circulation traces back to one moment. On 5 June 2025, CEO Aravind Srinivas spoke onstage at Bloomberg’s Tech Summit.
Srinivas said Perplexity received 780 million queries in May, and that the product was growing more than 20% month over month, as reported by TechCrunch the same day. In the same quoted passage, he described the trajectory from 3,000 queries on the first day in 2022 to 30 million queries a day at the time of the interview.
Those two numbers do not describe the same volume.
| Calculation | Result |
|---|---|
| 780 million queries divided by 31 days in May | 25.2 million per day |
| 30 million per day multiplied by 31 days | 930 million per month |
| Gap between stated daily rate and stated monthly total | 19.2% |
Both figures came from the same speaker, at the same event, in the same paragraph. One of them is rounded, or the daily figure is a June number attached to a May total, or the two metrics count different things. The transcript does not say which.
Roughly two hundred articles have since republished both numbers side by side without noting that they disagree. Any downstream calculation built on the 30 million daily figure runs 19% higher than the same calculation built on the 780 million monthly figure.
Figure 2. The two usage figures stated in a single passage at Bloomberg Tech Summit, 5 June 2025.
Fracture 2: The billion-query figure was a conditional forecast
The most widely republished Perplexity growth statistic is that the platform reached, or was on track for, a billion queries a week. The verbatim quote is narrower than the headline.
Srinivas said the company would be doing a billion queries a week within a year “if we can sustain this growth rate.” The conditional clause is the operative half of that sentence, and it is the half that gets dropped.
| Scale check | Value |
|---|---|
| One billion queries per week, daily equivalent | 142.9 million per day |
| One billion queries per week, monthly equivalent | 4.33 billion per month |
| Verified May 2025 monthly volume | 780 million |
| Multiple required to reach the target | 5.6x |
Reaching that target required 20% month-over-month compounding to hold for twelve consecutive months. No company disclosure confirms it was reached. The figure circulates as a current operating statistic anyway.
The same problem applies to the growth rate itself. Compounding 780 million at 20% per month for twelve months produces approximately 7 billion monthly queries by May 2026. No source claims that number, because nobody actually believes the rate held. The rate is quoted anyway, detached from the arithmetic it implies.
Fracture 3: Query growth was engineered to decouple from user growth
At the same event, Srinivas explained why the Comet browser mattered to the metrics. His description is unusually direct about the mechanism.
He said that if people are in the browser, it is “infinite retention,” and that activity in the search bar, the new tab page, the sidecar and every page visited would produce “extra queries per active user, as well as seeking new users.”
That is a stated design intent to raise queries per user independently of how many users exist.
The consequence for anyone reading Perplexity statistics is direct. Query volume growth cannot be read as user growth. A browser that converts passive browsing into logged queries increases the numerator without touching the denominator. Both numbers can grow, and the query figure will grow faster by construction.
Comet went free in March 2026. Every query-volume comparison that spans that date is comparing two different products.
Fracture 4: 360 million people were handed the paid tier, then it was withdrawn
In July 2025, Perplexity partnered with Bharti Airtel, India’s second-largest telecom operator. TechCrunch reported that the deal offered a free 12-month Perplexity Pro subscription, normally worth $200, to all 360 million Airtel subscribers across mobile, Wi-Fi and DTH. Airtel confirmed to TechCrunch that the arrangement was exclusive, meaning no other Indian telco could offer Perplexity access to its subscribers.
The scale of that grant relative to every published user figure is the point.
| Comparison | Ratio |
|---|---|
| Airtel subscribers granted free Pro versus 45 million MAU | 8.0x |
| Airtel subscribers granted free Pro versus 30 million core-product MAU | 12.0x |
| Activation rate that would add 18 million users | 5% |
A 5% activation rate on that grant would add 18 million users, equal to 40% of the most-repeated 2026 MAU figure. No activation rate has been published. Every MAU number covering the period from July 2025 onward sits on top of a promotional cohort of unknown size.
Around February 2026, Airtel replaced the Perplexity Pro benefit with Adobe Express Premium. Reliance Jio, the larger competitor, bundles Google Gemini. Any user series spanning mid-2025 to mid-2026 therefore contains a distribution cohort that was switched on and then switched off, and no published figure separates it from organic signups.
The deal’s own headline number drifts inside the reporting. Business Standard published a headline citing 390 million users while its body copy said 360 million. A 30 million discrepancy inside a single article is larger than Perplexity’s entire verified core-product user base in April 2025.
Fracture 5: Queries per user is an artifact of the denominator
The most commonly cited Perplexity engagement statistic is that the average user runs roughly 17 queries per month. That figure is arithmetic, not measurement. It is 780 million divided by 45 million.
Hold the numerator constant and change only which MAU figure is used.
| MAU figure used | What it counts | Queries per user |
|---|---|---|
| 30 million | Core product, April 2025 | 26.0 |
| 45 million | Unspecified, most-repeated 2026 figure | 17.3 |
| 100 million | All products combined, April 2026 | 7.8 |
The same company, the same query total, a 3.3x spread in apparent engagement. An article describing Perplexity users as heavily engaged and an article describing them as casual can both be correct, and neither has to misreport a single input.
The numerator is also stale. The 780 million figure is from May 2025. It is being divided by MAU figures dated early 2026. The ratio compares a number from one year against a number from the next.
Figure 3. Queries per user per month, holding the numerator constant and varying only the MAU denominator.
Fracture 6: Independent measurement puts Perplexity below the reporting threshold
Company-stated figures and third-party panel estimates share a weakness. They measure the platform from the platform’s side. Network-level measurement counts what actually arrives at websites.
Cloudflare Radar publishes crawl-to-refer ratios based on traffic across Cloudflare’s entire customer base, comparing HTML page requests from a platform’s crawler against HTML page requests that platform refers. Cloudflare launched the metric on 1 July 2025.
Radar’s referral data for May 2026 produces a very different picture of scale than the MAU figures.
| Referral source | Share of search referrals |
|---|---|
| 87.5% | |
| TikTok | 3.26% |
| ChatGPT | 0.28% |
| Gemini, Claude and Perplexity, reported together as “other” | 0.18% |
| All four AI chatbots combined | 0.29% |
ChatGPT is the only AI source that clears Radar’s quarterly reporting threshold. Gemini, Claude and Perplexity each sit below it and are folded into an aggregate bucket. TikTok alone sends approximately 11 times the referral volume of all four AI chatbots combined.
That is not an argument that Perplexity has no users. It is an argument that user counts and referral volume are separate measurements, and that a platform can hold 45 million monthly actives while sending a share of web referrals too small to report individually.
Figure 4. Share of search referrals across Cloudflare’s network, May 2026.
The crawl-to-refer ratio for Perplexity itself illustrates how unstable these numbers are across measurement windows.
| Window | Perplexity crawl-to-refer ratio |
|---|---|
| Late March 2025 | Above 700:1 |
| Q1 2026 | Approximately 111:1 |
| April 2026 | 95:1 |
| Week of 25 May to 1 June 2026 | 190:1 |
| News and publications sites | 32.7:1 |
The same metric, the same company, a 21x spread. Cloudflare’s data is tied to specific time windows and specific site categories, and the ratios must not be averaged across them. They routinely are.
For context on the same measure, ClaudeBot ran at 11,122:1 in the week of 25 May 2026 and OpenAI’s GPTBot at 857:1, against Googlebot at roughly 5:1. Perplexity is the most reciprocal of the major AI crawlers by a wide margin. It is also the one measuring below the threshold at which Cloudflare reports referral share separately.
Why the numbers diverge: six measurement decisions
Every figure in this report is defensible. The choices that produce them are not disclosed in the headline, and they compound.
- Product scope. Core product MAU versus all products combined. This single choice moves the figure from 30 million to over 100 million.
- Unit of measurement. Monthly active users, monthly visits, lifetime downloads and daily queries are four different units. Three of them are routinely presented as user counts.
- Measurement side. Company-stated internal analytics versus third-party panel estimates versus network-level observation. Each sees a different portion of the same activity.
- Distribution cohorts. Bundled subscriptions granted through telecom and device partnerships enter the same counter as organic signups, and exit it when the deal ends.
- Time window. The verified query figure is from May 2025. The MAU figures are from early 2026. Ratios built from both are cross-period.
- Forward-looking statements. Targets stated with explicit conditions are republished without them.
Cloudflare’s own documentation names a further limitation that applies to every AI usage figure in this report. Native app traffic, meaning activity inside the Perplexity, ChatGPT and Claude mobile and desktop applications, is a measurement blind spot for network-level referral data. Panel data has the mirror problem, undercounting logged-in desktop activity. No single method sees the whole picture.
What this means for B2B SaaS marketing
The practical question is not how many users Perplexity has. It is what budget the channel justifies.
Perplexity is a citation surface, not a traffic channel
At an aggregate referral share below the threshold Cloudflare reports separately, no realistic B2B site should model Perplexity as a meaningful source of sessions in 2026. Teams that set session targets for it will miss them regardless of execution quality. Measure it as brand visibility and citation presence, and hold revenue expectations against the channels that actually carry volume.
Reciprocity is better than the alternatives, and that matters for crawl policy
Perplexity’s crawl-to-refer ratio runs one to two orders of magnitude better than Anthropic’s or OpenAI’s. Cloudflare’s robots.txt analysis finds PerplexityBot and ChatGPT-User appearing more often in allow rules than disallow rules, which is a reasonable position given the return. Blocking decisions should be made per-operator on measured ratios, not as a blanket AI crawler policy.
The measurement problem is the strategy problem
A marketing team that cannot say which denominator a number uses cannot forecast with it. The same discipline that separates 30 million from 100 million MAU separates a pipeline forecast that holds from one that does not. This is the core of how SERPsculpt approaches B2B SaaS search programmes, and the same principle runs through the SERPsculpt research library, including work on B2B customer retention statistics where definitional drift produces the same kind of spread.
For teams building the surrounding programme, these measurement questions belong in the same conversation as channel mix and positioning, covered in more depth in this breakdown of sales and marketing strategy for B2B SaaS.
Methodology
Figures in this report were traced to their originating source before inclusion. Statements made by Perplexity executives are attributed to the event and publication that reported them, with the conditional and temporal qualifiers preserved.
Where a widely circulated figure could not be traced to a retrievable primary source, it was excluded and recorded in the accompanying verification logbook rather than presented with a hedge. The 230 million MAU claim appearing on a GEO agency marketing page falls into this category. It is attributed in that source to company reporting in March 2026 with no link, and no corresponding company disclosure exists.
Derived percentages and ratios are shown with their inputs so any reader can reproduce or dispute them.
Sources
TechCrunch, 5 June 2025: Perplexity received 780 million queries last month, CEO says
TechCrunch, 17 July 2025: Perplexity sees India as a shortcut in its race against OpenAI
TechCrunch, 30 March 2025: Perplexity CEO denies having financial issues, says no IPO before 2028